Acceleration Study of Two-Stage and Deep-Learning Based Facial Direction Detection on GPU-Based Edge Device
Hua-Luen Chen, Kuan-Hung Chen, Yin‐Tsung Hwang, Chih‐Peng Fan · 2022 IEEE 4th Global Conference on Life Sciences and Technologies (LifeTech) · 2022
When an intelligent autonomous mover is active, the facial direction messages of pedestrians become important for the use of social-aware navigation in surrounding-crowds environments. To speed-up the performance of the two-stage processing based convolutional neural network (CNN) design for facial direction detection on the cost-effective GPU-based edge device, in this paper, three acceleration methodologies are applied, including TensorRT, Multithreading, and Overclocking technologies, to speed-up the frames per second (FPS) performance on NVIDIA Jetson Nano platform. Compared with the direct two-stage implementation without using accelerations, the accelerated implementation performs more than three times speed-up ratios for facial direction detections of pedestrians.